A CT-based Radiomics Model Study For Predicting The Recurrence Of Locally Advanced Non-Small Cell Lung Cancer After Radical Operation

Purpose:The survival of locally advanced non-small cell lung cancer(NSCLC)after radical surgery is severe,and the occurrence of postoperative recurrence cannot be accurately predicted at present.The purpose of this study is to combine preoperative CT image radiomics features and related clinical factors to establish and verify a comprehensive prediction model for predicting postoperative recurrence in patients with locally advanced NSCLC.Methods:This study retrospectively analyzed 124 patients with locally advanced NSCLC(stage ⅡB～ⅢB)undergoing radical surgery from January 2014 to November 2018 in the Shandong Cancer Hospital affiliated to Shandong University.The clinical and pathological data of the patients were collected.All patients underwent enhanced CT before operation.The patients were randomly divided into a training group and a validation group,with 100 people and 24 people respectively.The 3D-Slicer software was used to manually outline the region of interest of the primary tumor,and 724 radiomics features were extracted.The LASSO algorithm is used for feature screening,and 10 meaningful features are screened out,and then the radiomic score(Rad-score)is constructed.The Cox multivariate regression model was used to determine the clinical independent risk factors for recurrence.The clinical factor nomogram,radiomics nomogram and comprehensive nomogram were constructed separately from the training group data.The predictive performance of the model is evaluated by the consistency index(C-index),and the clinical value of the model is compared through the decision curve analysis.Results:A total of 71 people developed relapses(53 people in the training group and 18 in the verification group).Pathological stage is independent risk factors for recurrence after radical operation of locally advanced NSCLC.For the clinical factor nomogram,the C-index of the training group is 0.652(95%CI:0.599～0.706),and the C-index of the verification group is 0.524(95%CI:0.462～0.586).For the radiomics nomogram,the C-index of the training group and the verification group were 0.786(95%CI:0.741～0.832)and 0.619(95%CI:0.547～0.692),respectively.For the comprehensive nomogram,the C-index of the training group and the validation group were 0.819(95%CI:0.777～0.861)and 0.623(95%CI:0.548～0.698),respectively.Compared with the clinical factor model,the comprehensive model has significantly better predictive performance(p<0.0001 for the training group and p<0.0001 for the verification group).Compared with the radiomics model,the predictive performance of the comprehensive model is also improved to a certain extent.Nevertheless,this improvement was not statistically significant in the validation group(p=0.028 for the training group,p=0.76 for the validation group).At the same time,compared with the clinical factor model,the radiomics model has relatively better predictive performance(p<0.0001 in the training group,p<0.0001 in the validation group).The analysis of the calibration curve and the decision curve analysis also showed that the comprehensive nomogram surpassed the clinical factors or radiomics nomogram in predicting postoperative recurrence of locally advanced NSCLC.Conclusion:The predictive power of the comprehensive nomogram constructed based on the preoperatively enhanced CT image of the primary tumor radiomics features and pathological stage is significantly higher than that of the individual clinical factor nomogram and radiomics nomogram.The comprehensive nomogram is most suitable for predicting the recurrence of locally advanced non-small cell lung cancer patients after radical operation.